| In the group of organisms,such as fish group and bird group,the ability of perception and action of individuals is limited,but follow some simple behavior rules,it can coorperate to complete complex activities such as migration,foraging,nesting and against enemy.It shows the orderly self-organization and coordination behaviors at group level,which is called group behavior.Human beings have derived tank formation,nautical fleet,aircraft formation and so on by referring the group behavior of animals.In recent years,with the expansion of application demand,in order to solve the problems of low load and single function of UAV(Unmanned Aerial Vehicle),UAV group technology has been developed,and has been widely used in aerial photography,logistics,agriculture and forestry operations,electric patrol,military and other fields.On account of the UAV group is invented,it is necessary to monitor it with some means,which can be used to assist with the training and to supervise and counter the invading of UAV group.The imaging distance of UAV is long,the imaging area is small,it usually occupies only dozens of pixels in the image,which is easy to be interfered by noise and background.Moreover,UAV has strong maneuverability during flight.In the face of UAV group,the features among multiple UAVs are not obvious,so it is a great challenge to detect and track each UAV target.In this paper,the detecting and tracking technology of UAV group is studied.For small target detection of UAV,an improved local contrast detection algorithm is proposed,which solve the problems of the high time-consuming of local contrast method and high false alarm rate of directional gradient detection.First,use the multistage directional gradient detection to obtain the salient region.Then,the local contrast detection is carried out according to the salient region image.The experimental results show that the proposed method has higher detection accuracy in the complex sky background compared with the directional gradient method and local contrast method.In order to improve the accuracy of detection,the YOLOv5 network was used to detect the UAV targets,and the self-made data set was used to train the network.For tracking UAV group,a group target tracking method based on dual matching is proposed.First,using adaptive threshold circle detection to divide the multi targets into groups.Then,conduct multi-target tracking that consider group as individual,use Kalman filter and Hungary algorithm to match detections with trajectorise.Finally,proceed the dual match with unmatched detections and trajectorise,propose a trajectorise maching strategy for the situation that group separation,merging and movement crossover,finish the tracking of group target.Finally,an image processing platform used to group target detecting and tracking is designed.Hardware design include high speed interface and high speed processor.Select FPGA as the master control,which is used to manage the reception and transmission of image and use four piece of DSP TMSC6678 to run the detecting and tracking algorithm of group target.FPGA receive ultra-high resolution image through QSFP+ optical module,and transmit divided image to each DSP with SRIO.DSPs working on master-slave model and computing with multi-core parallel,which ensure the processing speed of ultra-high resolution image.In addition,develop a upper software based on Py Qt library,as the tracking system display interface. |
